Emotion Recognition from Twitter Comments Using Deep Learning

Two individual humans may only communicate effectively if they recognize expressed emotions. Similarly, recognizing emotions from expressed language could effectively improve human-machine and machine-human interactions in applications where knowing expressed emotions at a given moment is of great importance. This paper discusses the implementation of two deep learning models, a CNN-based architecture model that uses n-gram filters and an n-hidden layers LSTM model on MATLAB that aim at detecting six emotions: Anger, Fear, Joy, Love, Sadness, and Surprise, from a dataset of annotated Twitter comments available on Kaggle, while utilizing word2vec word embeddings that display semantic meaning. The implemented n-hidden layer LSTM model acquired a macro-F1 score of 0.8764 on the test instances of the used dataset.